IMAGE MULTIPLICITY DESCRIPTIVE POWER OF POINT CLOUDS ACCURACY ASSESSMENT

For each area, a detailed analysis was performed, concerning point patterns and density TA1 Bare and flat terrain, on the pit lower part TA2 Bare and flat terrain, on the pit upper part TA3 Bare terrain, 33° inclined TA4 Bare terrain, 70° inclined Table 6. Main characteristics of detailed test areas Figure 11 shows point patterns for test area 1 lower flat area and for scenarios 1 and 4. Points have in general a constant den- sity, but there are lines in which density is much higher. Figure 11. Point patterns for test area n. 1 and for scenarios 1 left and 4 right Things are different when the sloped area n. 3 is considered. Figure 12. Point patterns for test area n. 3 and for scenarios 1 left and 4 right As shown by Figure 12, points are denser and there are no visible patterns full resolution images would confirm that. To better investigate point density, a statistical analysis was per- formed: for each test area and each scenario, the correspondent points were extracted; then a robust plane was fitted and the local slope and height determined; they were used to calculate the local density, normalized density and image GSD ground sampling distance. Results are shown in Table 7. Table 7 shows results for test areas 1, 3 and 4, and for all the five datasets considered. Considering the top part of the table, it re- ports that area 1 has a slope of 0° and a GSD value of 1.853.70 cm, for nadir images; 1.85 cm is the resolution of the acquired images; 3.70 cm is the effective resolution EGSD, that of the images used for point cloud extraction, adopting the recom- mended High strategy. The Density column reports normalized density values; the Density ratio column shows the ratio between the normalized density and that of a regular mesh having square cells with size equal to EGSD; Spacing is the linear spacing of an equivalent same density regular mesh having square cells; Spacing ratio is the ratio between the previous column and the local EGSD. In other words, if a point cloud had the same density of image pixels, Density ratio and Spacing ratio would be 1; if the density is higher, the former parameter is above 1 and the latter is below. TA 1 Slope: 0° GSD: 1.853.70 cm Scenario Density [ptsqm] Density ratio Spacing [cm] Spacing ratio S1 916 1.26 3.3 0.89 S2 911 1.25 3.3 0.89 S3 1501 2.06 2.6 0.70 S4 1358 1.86 2.7 0.73 S5 893 1.22 3.3 0.90 TA 3 Slope: 33° GSD: 1.773.54 cm Scenario Density [ptsqm] Density ratio Spacing [cm] Spacing ratio S1 1102 1.38 3.0 0.85 S2 1195 1.50 2.9 0.82 S3 1688 2.11 2.4 0.69 S4 1709 2.14 2.4 0.68 S5 1047 1.31 3.1 0.87 TA 4 Slope: 71° GSD: 1.743.47 cm Scenario Density [ptsqm] Density ratio Spacing [cm] Spacing ratio S1 1302 1.57 2.8 0.80 S2 1456 1.75 2.6 0.76 S3 1820 2.19 2.3 0.68 S4 2048 2.47 2.2 0.64 S5 1042 1.25 3.1 0.89 Table 7. Detailed analysis of point density

7. IMAGE MULTIPLICITY

Some initial results were obtained for image multiplicity, that is, the number of images to which a certain point on the terrain is projected. Figure 13 shows image multiplicities for test area 1 3x3 metres and for the scenarios 1, 2, 3 and 5. The minimum and maximum values are, from top-left and clockwise: S1, 27 and 33; S2, 184 and 190; S3, 30 and 47; S5, 12 and 15. Such differ- ences depend on the number of the images used, see Table 4. A detailed and thorough study of image multiplicity and its corre- lation to quality parameters will be inserted in future contribu- tions. Figure 13. Image multiplicity for test area 1 and for the scenarios 1 top-left, 2, 3 and 5, clockwise This contribution has been peer-reviewed. The double-blind peer-review was conducted on the basis of the full paper. doi:10.5194isprsannals-III-1-175-2016 179

8. DESCRIPTIVE POWER OF POINT CLOUDS

It’s only a curiosity, but we were really impressed by the descrip- tive capability of the point clouds analysed and of the obtained maps. Figure 14 left shows the tire tracks of the dump trucks and bulldozers used in the pit. Figure 14 right demonstrates how well these tracks are detectable in the slope map; they are also visible in the density map, which is not shown here. Figure 14. Tire tracks and the correspondent part of the slope map

9. ACCURACY ASSESSMENT

Point cloud accuracy was assessed for all the scenarios with three sets of check points:  the 18 signalized points shown in Figure 6 CKP1;  a set of 118 points measured with a topographic total station on the upper flat area, shown in yellow in Figure 15 CKP2;  a set of 170 points measured with a topographic total station on the scarp, shown in red in Figure 15 CKP3. Figure 15. The control points used for accuracy assessment A functionality was developed of the Toolbox Section 5, which is able to compare the check points against the surface described by point clouds. The 3D distance was formed for any point and any scenario considered and then results were summarized in terms of the RMSE of the distance. A moderate blunder detection was performed. Check point total number inlier number RMSE [cm] S1 S2 S3 S4 S5 CKP1 1818 4.3 1818 2.1 1818 3.6 1818 2.4 1818 7.6 CKP2 117118 3.3 117118 2.5 117118 3.4 117118 3.6 118118 7.6 CKP3 160170 5.7 161170 6.6 159170 5.3 160170 6.1 161170 6.0 Table 8. Accuracy assessment for the five scenarios considered and for the three check point sets Table 8 reports accuracy results for all the five scenarios on the columns and the three check point sets on the rows. Considering for istance the crossing between row CKP3 and column S1 , 160170 means that results concern 160 check points out of 170 and the RMSE value for the 160 3D distances is 5.7 cm. Figure 16. 3D distances between CKP2 set and the scenario 1 Our developed toolbox can perform different types of visualization. Figure 16 shows the 3D distances between the CKP2 set 118 points on the upper flat area and scenario 1. Quality decay is clearly visible for points close to the border of the test site; that is reasonibly originated by local block weakness and low image multiplicity. Figure 17. 3D distances between CKP2 set and the scenario 2 Scenario 2, based on different imagery and different geometry, clearly has another behaviour, see Figure 17. Figure 18. 3D distances between CKP2 set and the scenario 5 Finally, scenario 5 has the same behaviour as 1, as it has North- This contribution has been peer-reviewed. The double-blind peer-review was conducted on the basis of the full paper. doi:10.5194isprsannals-III-1-175-2016 180 South strips in common, but is weaker, due to the lack of cross strips, see Figure 18, Figure 4 and Figure 5.

10. DISCUSSION